Native Signal Processing

Intel and other processor vendors have championed native signal processing (NSP). The concept is simple: Given ever-more-powerful processors, the host should be able to take on a longer list of multimedia chores. The first NSP implementation appeared on the PowerPC-based Macintosh from Apple Computer in March 1994. The host PowerPC 601 has signal-processing support for speech, telephony, and many audio/video functions as well.

How much can the CPU do? To implement NSP, Intel recommends a baseline target platform that includes an audio codec and an accelerated graphics chip, thereby leaving the Pentium free to perform DSP functions. Without the graphics-acceleration chip, the decode and playback of an Indeo video clip consumes 100 percent of a 90-MHz Pentium's bandwidth. By off-loading BITBLT and color-conversion operations to the graphics chip, 60 percent of the Pentium processor becomes available.

Perhaps the most dramatic example of NSP is Vivo Software's (Waltham, MA) H.320-compliant codec package, in which a Pentium host does all the compression/decompression, multiplexing, and control functions for a videoconference. Other NSP demonstrations include wave-table synthesis, digital mixing, sample rate conversion, MIDI synthesis, and A/D pulse-code modulation compression/decompression.

In general, compute-intensive tasks and heavily interrupt-driven applications for audio, video, and telephony (e.g., MPEG-2, G.728, and V.34 modulation) benefit from DSP hardware in the NSP environment. Hence, under the hood, most NSP-labeled Pentium platforms will include several DSP engines dubbed smart codecs, modem data pumps, and graphics accelerator chips.

Intel is no t standing still, however. The upcoming P55C chip is a Pentium with greatly enhanced integer performance and will be better suited for multimedia algorithms. Intel also is reworking its NSP plans to function within Microsoft's approach to NSP, which is based on DirectX. As host platforms migrate from Pentium to Pentium Pro to P7 technology, the ability to perform more real-time signal-processing and multimedia functions increases. This assumes, of course, that the OSes and user-friendly applications of the future do not expand to consume all available processing power.